MANARمنار
SOVEREIGN INTELLIGENCE · BUILT ON YOUR GROUND hello@manar.pk →
Get in touch

Open a conversation.

Tell us what you run and what sits unread. We will tell you which model reads it, and reply within two working days.

{{ cError }}

Your details stay with Manar. No data leaves, on the website or on your ground.

Your inquiry is on its ground.

Thank you, {{ cName }}. We have your message and will be in touch within two working days.

Manar · Sovereign intelligence

Not the AI that knows the world The AI that knows yours.

Trained on your domain. Running on hardware you own. No cloud, no data leaving your site.

Site Read The thesis How adoption works
Pakistan's AI Adoption Partner منارة في الظلام
What we do

Your data. Your walls. Your intelligence.

Every organization sits on data it has never read. Policies no one has time to search. Cameras no one has time to watch. The data is there. It sits unused.

Manar reads it. We train a registered model on your specific conditions, install it on hardware you own, and let it compound sharper every day it operates. The intelligence never leaves your walls.

The premise
Generic beats no one.

Generic AI is available to everyone, which is precisely why it is the advantage of no one. The advantage is sovereign intelligence, trained on your conditions alone. Stanford's AI Index: the cost of generic intelligence fell 280 fold in eighteen months. What everyone can rent is an advantage to no one.

The approach
Built to extend.

A new kind of work gets a model trained for it, not a system rebuilt from nothing. What it learns about your operation stays with you.

The company
One method. Seventeen models.

Manar trains and deploys registered models, on hardware sized to what each one does, inside your environment. Built to grow.

What we believe
01
Specific beats general. Always.

Your site. Your data. Your conditions. Generic AI was not trained on any of them.

02
Starting is the advantage. Waiting is the cost.

Every month you train builds a record no competitor can replicate on any budget or timeline.

03
The moat is the data. Not the model.

Models can be copied. A proprietary record of your environment, built over years, cannot be reproduced.

04
Your roof. Your architecture. Your advantage.

Running in your building is not a lesser version of the cloud. It is the sovereign version.

05
Adoption is a build. Not a subscription.

What is rented visits your operation. What is built and owned becomes part of it. Manar engagements end with assets on your books and capability in your staff, or they have not ended.

How adoption works →
The record

Adoption is failing.
The causes are known.

The public record on generic AI adoption is now measured: most pilots return nothing, abandonment is rising, and the obstacles organizations name most are cost, data privacy, and security. Not one of these is a property of the intelligence. All of them are properties of how it was adopted: generic, rented, and cloud bound. Manar runs adoption the other way, in five stages, ending in a system you own, run by your staff, that met a number before you accepted it.

The Adoption Path →
Where it fits

Built for what you run. Not the world in general.

Every sector on this list has been running data for years that nobody has read. Manar trains a model on your specific conditions and deploys it on hardware you own, inside your building.

Institutions
AI that reads your archive. Policy, records, and registers answered in seconds.
Security
AI that watches your cameras. Trained on your site, not a generic dataset.
Factories
AI that reads your production line. Defects caught before they move downstream.
Oil and Gas
AI that monitors your field. Telemetry read on-site, offline, no cloud required.
Agriculture
AI that reads your land. Decades of yield history made useful for the first time.
Healthcare
AI that reads your records. Patient history answered in seconds. No data leaves the building.
All nine sectors →
Where it runs

It runs on what you already have. On hardware sized to the job.

A deployment does not rebuild your site around it. The model plugs into the systems you already run, on hardware sized to the job, and works with what they already produce.

CONNECT

What you already run

The cameras, machines, records, and software already running at a site. A registered model works with the systems operating today. It does not replace them.

Cameras · Machines · Records · Software
READ

Live data in place

The model reads data where it already lives. No export, no transfer to a third party, no rebuild of the operation around our software.

No export · No transfer · No third-party access
REASON

Local intelligence

Reasoning happens on site, on hardware sized to the model deployed there, owned by the client or specified by Manar. Decisions, alerts, and answers delivered where the work happens.

On-site hardware · Client-owned · Sized to workload
REMAIN

Sovereign throughout

None of this requires data to leave the building. Integration extends what a model can read. It does not change where the intelligence runs or who owns it.

Data stays inside · Sovereign from day one
How we reach further

Three ways in. Your data never moves.

A partner often already has the infrastructure, the install base, or the software a client uses every day. Manar trains the model behind it. However a deployment starts, the client's data stays in their building.

Direct

We install it ourselves

Manar trains the model, ships it, and installs it inside the client's environment. We hold the relationship from the first conversation to ongoing support.

Through an installer

They already hold the relationship

A systems integrator or distributor already sells and maintains the hardware a client runs. Manar trains the model for that hardware. They keep the client and the install. We keep the model.

Inside their product

They already hold the software

A partner's existing software calls the model as a feature inside their own product. Their customers see the partner's interface, not ours. We supply and train the model behind it.

Partnership paths →
How we work

From the registry. Onto your ground.

A deployment starts with one of the registered models. It gets trained on your data, installed on your hardware, and kept current as your operation changes.

01 Train

Start from a registered model already built for the classification a client needs. Train it further on their own data until it knows their specific conditions.

02 Deploy

The trained model installs on hardware the organization owns. In its building, behind its walls, offline if required. No cloud dependency. No data leaves.

03 Handover

Staff who will work with it are trained before engagement closes. Complete when your own staff operate the system without Manar in the room.

04 Adapt

The model retrains as the client's own data grows, so it reflects the operation as it is today, not as it was on day one.

05 Compound

Accuracy on a client's own conditions grows with every day the model runs there. Holds a record no later competitor can shortcut.

Full methodology →
How a model is built
HIKMAحكمة

The House of Wisdom took what existed and refined it into something that did not exist before. Hikma works the same way. We take open foundation models and teach them what they have never encountered: local conditions and the data of your specific domain.

What we start with

The strongest open foundations available.

The open frontier advances every month. We begin with the strongest available foundation for each domain. The starting point is public. What it becomes is not.

What we teach it

Your data. Conditions no global training has seen.

Each foundation learns the data of the domain and environment it will serve. Local conditions. Local patterns. The record no competitor can collect at any speed.

What you own

Runs on your hardware. Belongs to no one else.

Refined intelligence deploys on hardware the organization owns, behind its own walls, offline if required. Sovereign from the moment it leaves Hikma. Proprietary from the first inference.

How HIKMA transforms intelligence
What enters
Open foundations.
Available to everyone.
The strongest models in the world are public. That is where we begin. Where everyone begins.
حكمةHIKMA
What leaves
Proprietary intelligence.
Trained on your world alone.
Intelligence that has learned your data, your conditions, your patterns. Sovereign. Offline. It belongs to you.
VISION · LANGUAGE · TIME SERIES · EARTH OBSERVATION
YOUR DATA · YOUR CONDITIONS · YOUR INTELLIGENCE · YOUR WALLS
Browse full registry →
Common questions

Before you ask

Who owns the model once it is built? +

You do. The model and the hardware it runs on belong to you from the moment installation completes. Nothing is licensed back, and nothing depends on Manar continuing to operate for you to keep using what was built.

How do we know it actually stays offline? +

You check it yourself. At handover, the network connection is physically removed while the system is answering a live question, in front of your own staff, and it keeps answering. Full detail is on the security page.

We tried AI before and it did not stick. Why would this be different? +

Because what you experienced has now been measured across thousands of organizations, and the causes are known: the tools were generic, rented, and cloud bound. The Adoption Path reverses each one, and a deployment is accepted only when it meets a number agreed on your conditions before work begins. The full record is on the adoption page.

View all questions on FAQ page →